Recognition of pomegranate on tree and stereoscopic locating of the fruit

Abstract

Mechanized harvesting of pomegranate is still a challenge due to tangled shape of the tree. Robotic harvesting seems to be the final solution in this case. The first step for this aim is to recognize the pomegranate fruit and locate its spatial position on the tree which has been the objectives of this approach. Color features of three different color spaces, RGB, HSV and YCrCb were studied to discriminate the fruit from the surrounding objects in the images. Two-dimensional locations of the fruits were determined using their corresponding centroid in the images. Stereoscopic images captured from two identical cameras were used to determine the depth of the ROIs in the images. This would yield the 3D location of the fruits which is needed for the harvesting robot to pick the fruit. Results showed that Cr component of the YCrCb color space has been the best criterion for segmentation of fruit from the other objects in the images with 100% correct recognition rate. Maximum distance estimation error (DEE) of 2.4 cm was determined for the stereoscopic vision system which shows its consistency to be used as a recognition and locating tool for pomegranate harvesting robot.